We investigate advanced methods for improving security and preventing intrusions in logistics hubs and ports. For this purpose, we employ unmanned aerial vehicles (UAVs) and object detection artificial intelligence (AI) methods. In recent years, UAVs have become an increasingly popular tool for surveillance due to their ability to provide real time video feed from difficult or inaccessible areas. Leveraging these UAV strengths, we use the YOLOv3 algorithm for real time object detection in surveillance images captured by UAVs. We train the YOLOv3 algorithm using various combinations of three UAV-generated datasets; VisDrone, DAC-SDC and the Stanford Drone Dataset. We then test the model versions obtained from each training dataset on independent images created under practical conditions. The results demonstrate the importance of robust training by using a dataset that has similar characteristics with the environment in which the system will operate.